Forum Discussion

Joel_Pinto_1689's avatar
Joel_Pinto_1689
Icon for Nimbostratus rankNimbostratus
Sep 04, 2014

Issue with GTM/LTM SNMP udp_gtm_dns

Hi, Im having the following problem with 4 big ip 1600 (currently running GTM) This 4 boxes are working as two separate clusters active/standby. I have a cacti learning with snmp the amount of total connections and it tells me that i have 8 millions in GTM-A-1 and in the other cluster GTM-B-1 (hipotetical names) i have 10 thousands of conections, always talking about the active server. The standby servers also give me via snmp info related to conections (like 3 or 4 alone connections, i assumed this was a bug)

 

Important fact: When i enter the dashboard of the GTMs they tell me something very diferent like no Open Conections, zero. I dont have this issue with another cluster of BigIP 2000s (Newer Version, same service and isolated from these other 2 clusters that are giving me a headache.) for the same service.

 

All of this as UDP Open conections - udp_gtm_dns on my cacti.

 

(They are resolving querys for mobility in 3G, Gn interface)

 

Then, to sumarize: I have two clusters and one of them have 8 million connections while the other doesn't have any related to that one. They are in the same GTM Group (replicating zones)

 

The partner that sold the appliances told me that i should boot over the two clusters but im searching for a different approach. Maybe something is slipping from me in cacti in method of how it graphics

 

That with the GTMs.

 

Now, with LTMs (Big IP 4200v) i have a similar issue. I have 2 Clusters balancing DNS querys to two different DNS servers. By datasheet, each appliance have a maximum 2.5 million L4 http Request per second, and 10 million maximum concurrent conections. Past month, with cacti i had 88k QuerysPerSecond (in my DNS servers) and 3.9millions conections in the LTM (using the same object as with the GTMs: UDP Open Conections udp_gtm_dns) This month, i have 135 kqps (growing fast) and 4.1million conections. There is something that im definitely doing wrong here or all my different appliances from f5 have issues with udp "conections".

 

Did somebody had the same issue? Thanks in advance! Joel.

 

application delivery
LTM
management
performance
snmp

5 Replies

Sort By
  • nitass's avatar
    nitass
    Icon for Employee rankEmployee
    Sep 05, 2014

    Im having the following problem with 4 big ip 1600 (currently running GTM) This 4 boxes are working as two separate clusters active/standby.

     

    are they running the same version? what are they?

     

    This month, i have 135 kqps (growing fast) and 4.1million conections.

     

    is 4.1M Total Connections?

     

  • Joel_Pinto_1689's avatar
    Joel_Pinto_1689
    Icon for Nimbostratus rankNimbostratus
    Sep 05, 2014

    Hi, Yes, The four GTMs: they all are running 10.2.4 Build 595.0 Hotfix HF3

     

    And for LTM i have 3.5M Total Open Conections in Cacti via SNMP (Right Now) In the dashboard of my LTMs i see another number that varies between 2.1 and 2.6M very unusual and rarely it goes up to 2.7M (Right now too.) Always talking about Open Conections. The New Conections field tells me i have 39K per second conections.

     

    Here's a screen cap of the dashboard at this very moment.

     

     

    Thanks in advance! Joel.

     

  • nitass's avatar
    nitass
    Icon for Employee rankEmployee
    Sep 06, 2014

    And for LTM i have 3.5M Total Open Conections in Cacti via SNMP (Right Now) In the dashboard of my LTMs i see another number that varies between 2.1 and 2.6M very unusual and rarely it goes up to 2.7M (Right now too.) Always talking about Open Conections. The New Conections field tells me i have 39K per second conections.

     

    does active connections performance graph (overview / performance) match cacti?

     

    i remember that dashboard realtime graph uses different data source. anyway, 3.5M and 2.6M is quite significant.

     

  • Joel_Pinto_1689's avatar
    Joel_Pinto_1689
    Icon for Nimbostratus rankNimbostratus
    Sep 08, 2014

    Yes, Matches. But All related to performance (cpu, memory utilization) matches cacti. The only trouble is the one with the conections.

     

    The problem is that i think the cluster is at max usage. By datasheet it supports the following:

     

    L7 requests per second: 850K

     

    L4 connections per second: 300K

     

    L4 HTTP requests per second: 2.5M

     

    Max Concurrent Conections: 10M

     

    Here, some caps related to performance:

     

    Active (Open?) Conections in the F5 Perfomance/Overview ]

     

     

    Open Conections in Cacti

     

    New Conections in Cacti (Matches with performance dashboard.)

     

    Implying im only balancing DNS, im at maximum usage in L4 Requests? 2.5millions is the number by datasheet.

     

    If im in high capacity ocupation/near top ocupation im going for a Viprion. The thing thats bugging me its the fact that the partner that sold me this bigIPs dont have the smallest idea about this issue or how im supossed to do the measurements of the platform really gets me.

     

    Anyway, do you now if the OID/datasource of the realtime dashboard is available? I cant find it in the MiBs!

     

    Thanks in advance for answering!

     

  • Joel_Pinto_1689's avatar
    Joel_Pinto_1689
    Icon for Nimbostratus rankNimbostratus
    Sep 11, 2014

    For giving closure to the question:

     

    I spent a few nights and find out that i wasn't polling the virtual servers and instead i was polling the summary of udp conections. Therefore, the cacti was showing me client and server side udp conections, and to find this value in the appliances i went here: (Module statistics > Profile summary > UDP )

     

     

    Also, i managed to polish the dashboard open conections in realtime by aplying load balancing udp datagram on that profile so that it doesn't show me mixed tcp and udp and only show me udp sessions.

     

    In clean, the virtual server didn't have the profile i was polling so i was polling by default all udp in my LTMs.

     

    ps: The maximum of concurrent conections is by datasheet of 10 million. I found out that it is in fact 20Million because it sumarizes client side conections and server side conections. (in+out) (Confirmed by my Partner)

     

    Anyway, Thanks for your help Nitass! Best regards, Joel.